1. 1 Statistical Mechanics and Multi- Scale Simulation Methods ChBE 591-009 Prof. C. Heath Turner Lecture 20 Some materials adapted from Prof. Keith E. Gubbins: http://gubbins.ncsu.eduhttp://gubbins.ncsu.edu Some materials adapted from Prof. David Kofke: http://www.cbe.buffalo.edu/kofke.htmhttp://www.cbe.buffalo.edu/kofke.htm

2. 2 Truncating the Potential  Bulk system modeled via periodic boundary condition not feasible to include interactions with all images must truncate potential at half the box length (at most) to have all separations treated consistently  Contributions from distant separations may be important These two are same distance from central atom, yet: Black atom interacts Green atom does not These two are nearest images for central atom Only interactions considered

3. 3 Truncating the Potential (for MD)  Potential truncation introduces discontinuity Corresponds to an infinite force (f = -dU/dr) Problematic for MD simulations ruins energy conservation  Shifted potentials (force is unaffected) Removes infinite force Still discontinuity in force  Shifted-force potentials Routinely used in MD  For quantitative work need to re-introduce long-range interactions

4. 4 Truncating the Potential  Lennard-Jones example r c = 2.5 

5. 5 Radial Distribution Function  Radial distribution function, g(r), (p. 310 of Leach) key quantity in statistical mechanics quantifies correlation between atom pairs  Definition  Here’s an applet that computes g(r) Here’s an applet Number of atoms at r for ideal gas Number of atoms at r in actual system Q: What would g(r) of an ideal gas look like? Of a solid?

6. 6 Properties of Radial Distribution Functions  Internal Energy  Pressure  Pair distribution function can be derived from experimental X-ray diffraction (Fourier transform of structure factor, S(k))  Long-range corrections VERY important for phase equilibria predictions.

7. 7 Simple Long-Range Correction  Approximate distant interactions by assuming uniform distribution beyond cutoff: g(r) = 1 for r > r cut  Corrections to thermodynamic properties Internal energy Virial Chemical potential Expression for Lennard-Jones model For r c /  = 2.5, these are about 5-10% of the total values (depends on density and conditions)

8. 8 Coulombic Long-Range Correction  Coulombic interactions must be treated specially very long range 1/r form does not die off as quickly as volume grows finite only because + and – contributions cancel  Methods Full lattice sum Here is an appletHere is an applet demonstrating direct approach Ewald sum Treat surroundings as dielectric continuum

9. 9 Aside: Fourier Series  Consider periodic function on - L/2, +L/2  A Fourier series provides an equivalent representation of the function  The coefficients are One period

10. 10 Fourier Series Example  f(x) is a square wave

11. 11 Fourier Series Example  f(x) is a square wave n = 1

12. 12 Fourier Series Example  f(x) is a square wave n = 1, 3, 5, 7

13. 13 Fourier Representation  The set of Fourier-space coefficients b n contain complete information about the function  Although f(x) is periodic to infinity, b n is non-negligible over only a finite range  Sometimes the Fourier representation is more convenient to use

14. 14 Observations on Fourier Sum  Smooth functions f(x) require few coefficients b n  Sharp functions (square wave) require more coefficients  Large-n coefficients describe high-frequency behavior of f(x) large n = short wavelength  Small-n coefficients describe low-frequency behavior small n = long wavelength e.g., n = 0 coefficient is simple average of f(x)

15. 15 Fourier Transform  As L increases, f(x) becomes less periodic  Fourier transform arrives in limit of L    Compact form obtained with exponential form of cos/sin  Useful relations derivative convolution inverse forward a n = real part of transform b n = imaginary part

16. 16 Fourier Transform Example  Gaussian  Transform is also a Gaussian!  Width of transform is reciprocal of width of function k-space is “reciprocal” space sharp f(x) requires more values of F(k) for good representation  (x-x o ) transforms into a sine/cosine wave of frequency x o :

17. 17 Fourier Transform Relevance  Many features of statistical-mechanical systems are described in k-space structure transport behavior electrostatics  This description focuses on the correlations shown over a particular length scale (depending on k)  Macroscopic observables are recovered in the k  0 limit  Corresponding treatment is applied in the time/frequency domains

18. 18 Ewald Sum  We want to sum the interaction energy of each charge in the central volume with all images of the other charges express in terms of electrostatic potential the charge density creating the potential is this is a periodic function (of period L), but it is very sharp Fourier representation would never converge + + + - - - + + + - - - + + + - - - + + + - - - + + + - - - + + + - - - + + + - - - + + + - - - + + + - - - + + + - - - + + + - - - + + + - - - + + + - - - + + + - - - + + + - - - + + + - - - + + + - - - + + + - - - + + + - - - + + + - - - + + + - - - + + + - - - + + + - - - + + + - - - + + + - - -

19. 19 Ewald Sum  The charge-charge interaction between the charges in the central box and all images of all particles in the surrounding boxes is:  There are contributions within the central box + interactions with surrounding boxes. There is also a contribution between the group of surrounding boxes and the surrounding medium.  Problems: summation is conditionally convergent, varies rapidly at small distances, converges very slowly.  Solution:

20. 20 Ewald Sum  In the Ewald method, each charge is assumed to be surrounded by a neutralizing charge distribution of equal magnitude (but opposite sign). A Gaussian charge distribution is used: include n = 0 Large  takes  back to  function The potential is given as: This sum converges rapidly. Only the screened charge from each particle interacts with the periodic images.

21. 21 Ewald Sum  A second sum is then added to exactly counteract (correct) the neutralizing distribution. This compensating distribution is performed in reciprocal space (k). These reciprocal vectors are: k = 2  n/L: This reciprocal space sum also converges much more quickly. Must find balance of specified parameters: Width of Gaussian Number of reciprocal vectors

22. 22 Ewald Sum  A third term must also be included to account for self-interaction and subtracted from the total:  A fourth correction term accounts for the surrounding medium. This depends on the relative permitivity. It is usually assumed to be infinite (a conductor) and no correction is needed. If the surrounding medium is a vacuum then a correction must be added:

23. 23 Ewald Method. Comments  Basic form requires an O(N 2 ) calculation efficiency can be introduced to reduce to O(N 3/2 ) good value of  is 5L, but should check for given application can be extended to sum point dipoles  Other methods are in common use reaction field particle-particle/particle mesh (better for larger systems) fast multipole (better for larger systems)

24. 24 Summary  Contributions from distant interactions cannot be neglected potential truncated at no more than half box length treat long-range assuming uniform radial distribution function  Coulombic interactions require explicit summing of images too costly to perform direct sum Ewald method is more efficient smear charges to approximate electrostatic field simple correction for self interaction real-space correction for smearing